NVIDIA H100 vs Google TPU v7 Ironwood
Google TPU v7 Ironwood comes out ahead of NVIDIA H100 on 6 of the 6 things that decide this, going by what each maker publishes. The one that matters for you may be one the leader loses.
The verdict
Google TPU v7 Ironwood takes it, 6–0 on rounds.
On the published figures, Google TPU v7 Ironwood takes 6 of 6 rounds. Treat that as a starting point and check the current figures on both pages, because prices and plans move.
Choose NVIDIA H100 if…
- • You have a reason to pick NVIDIA H100 that is not on this list
Choose Google TPU v7 Ironwood if…
- • You want Google TPU v7 Ironwood's edge on how much memory it has and how fast its memory is
Round by round
- Round 1How much memory it hasGoogle TPU v7 Ironwood wins
Google TPU v7 Ironwood wins this round on what the company itself publishes.
- Round 2How fast its memory isGoogle TPU v7 Ironwood wins
Google TPU v7 Ironwood wins this round on what the company itself publishes.
- Round 3How fast it runs FP8 AI mathGoogle TPU v7 Ironwood winsNVIDIA H1005/10
1,979 TFLOPS
fp8 tensor core
- 1,979 TFLOPS, against 4,614 TFLOPs
Source: dam-cdn.nvd.orangelogic.com ↗
Google TPU v7 Ironwood wins this round on what the company itself publishes.
- Round 4How fast it runs FP16 AI mathGoogle TPU v7 Ironwood winsNVIDIA H1005/10
989 TFLOPS
fp16 tensor core
- 989 TFLOPS, against 2,307 TFLOPs
Source: dam-cdn.nvd.orangelogic.com ↗
Google TPU v7 Ironwood8/102,307 TFLOPs
BF16 peak per chip (Google publishes no FP16 figure)
- peak bf16 compute per chip
Source: docs.cloud.google.com ↗
Google TPU v7 Ironwood wins this round on what the company itself publishes.
- Round 5Which memory generation it usesGoogle TPU v7 Ironwood wins
Google TPU v7 Ironwood wins this round on what the company itself publishes.
- Round 6How new it isGoogle TPU v7 Ironwood wins
Google TPU v7 Ironwood wins this round on what the company itself publishes.
Side by side
- peak bf16 compute per chip
- 80 GB HBM3, against 288 GB HBM3E
- 3.35 TB/s, against 8 TB/s
- 1,979 TFLOPS, against 5 PFLOPs
- 989 TFLOPS, against 2.5 PFLOPs
- 192 GB HBM, against 288 GB HBM3E
- 7.37 TB/s, against 8 TB/s
- 4,614 TFLOPs, against 5 PFLOPs
- 2,307 TFLOPs, against 2.5 PFLOPs